ZFS WD Red 4k alignment

[Gherardo]

Hello everyone,

I have a FreeBSD 9.2 machine that I use mostly as a file server/repository for my network. I'd like to upgrade the storage and I was thinking of using three or four WD30EFRX in RAID-Z1, but I'm not sure how to correctly format 4k drives (maybe it's not an issue anymore).

If I simply plug the drives and zpool create tank raidz1 ada0 ada1 ada2 will these three drives be automatically aligned? Or do I have to use gnop and align on every single drive? If the solution is the latter, does this look right?

gnop create -S 4096 /dev/ada0
gnop create -S 4096 /dev/ada1
gnop create -S 4096 /dev/ada2
zpool create tank raidz1 /dev/ada0.nop /dev/ada1.nop /dev/ada2.nop
zpool export tank
gnop destroy /dev/ada0.nop /dev/ada1.nop /dev/ada2.nop
zpool import tank

Thanks in advance for any input,

Gherardo

 

[Beeblebrox]

You first align the partition: gpart add -t freebsd-zfs -a 4k -s <size> <geom>. Then you must use gnop() to align ZFS to 4k. Once you have created the zpool that has ashift=12, adding partitions/HDDs to the zpool will always result in ashift=12. This means you can do it either way - create a zpool on one partition then add other partitions or create a zpool with all partitions in one go.

So yes, your code looks right.

EDIT: There are arguments both pro and con for giving ZFS the entire HDD. I'm of the opinion that partition first, then ZFS is healthier but as I stated it's arguable.

 

[Gherardo]

Thanks for the reply. I just completed the procedure and everything went fine

On a side note (maybe it's intended) I noticed a weird thing: if you just create a GPT label and then use gnop to set the geometry, the disk reverts to MBR (although it's temporary).

• gpart create -s gpt ada0 creates a GPT disk as it should.
• gnop create -S 4096 ada0 the disk is now aligned but gpart show says it's MBR!
• After you gnop destroy /dev/ada0.nop it reverts back to GPT, but at that point I had already created the RAID-Z1 pool.

So if you export the pool before deleting the *.nop it gives you an error saying that the GPT table is corrupted (and it really is, being MBR).

Anyway, thanks again!

 

[jem]

Some disks now report their 4K sector sizes using the DIOCGSTRIPESIZE ioctl. See the output of diskinfo -v /dev/adaX to see if yours does:

# diskinfo -v ada4
ada4
        512             # sectorsize
        2000398934016   # mediasize in bytes (1.8T)
        3907029168      # mediasize in sectors
        4096            # stripesize
        0               # stripeoffset
        3876021         # Cylinders according to firmware.
        16              # Heads according to firmware.
        63              # Sectors according to firmware.
        S2H7J90B308992  # Disk ident.

ZFS automatically creates pools with an ashift value of 12 on such disks.

If you build a pool on whole disks, without any underlying partitions, then it will be 4K sector aligned by default as it starts on sector 0.